Chainguard launches Harness for AI-driven vulnerability discovery

Harness finds code flaws after scanners pass

Chainguard announced Harness on Wednesday, an AI-powered security harness that uses frontier models, including Anthropic's restricted Claude Mythos 5.1, to find and fix exploitable vulnerabilities in application code that conventional scanners miss. Developers can use Harness without direct access to Mythos, but Chainguard has not disclosed which models handle vulnerability discovery for those customers. The company said Harness has found serious vulnerabilities in codebases that already had extensive security scanning, including software used in critical infrastructure. It did not identify the projects, flaws, or remediation status.

Workspaces and Checks move agent execution under policy

The launch pairs Harness with Workspaces, Checks, and expanded Libraries protections, extending Chainguard beyond hardened container images and dependencies into agent execution environments and CI verification. Workspaces gives each developer and coding agent a hardware-isolated microVM, with network policies that limit access to internal systems and credentials kept outside the agent environment. Signed activity logs record what each agent did for audit trails across autonomous sessions. Harness will also be available through Anthropic's Claude Marketplace, so teams can use existing Anthropic spending commitments instead of securing separate budget.

Checks lets coding agents run CI checks during development and receive signed results in seconds. Agents can run tests, inspect failures, and revise code before opening a pull request, avoiding CI queues as coding agents become shared team infrastructure running multiple sessions in parallel. It uses warm snapshots with dependencies already installed, refreshing them in the background when the lockfile changes. Each workflow step runs in a separate hardware-isolated microVM built from Chainguard OS packages, including FIPS-compliant toolchains. Runners do not hold long-lived credentials. A host-side proxy injects short-lived credentials into outbound requests, while declarative egress policies restrict network access and can end a microVM that attempts an unauthorized connection. Successful runs produce signed attestations tied to the execution environment, test results, and commit for merge gates to verify without rerunning the suite. Checks is in beta now.

Survey and Libraries expansion

Chainguard also released The AI Exploit Gap, a Wakefield Research survey it commissioned of 1,300 security leaders across nine markets. More than half said their organizations knowingly leave high or critical vulnerabilities unpatched at least sometimes. The figure was 66 percent at companies with $1 billion or more in revenue and 27 percent at companies below $250 million. Chainguard CISO Quincy Castro said AI can find and chain flaws in minutes, while fixes still move at human speed and compete with product deadlines and uptime.

Libraries is adding secure package endpoints for Go and .NET, plus CVE remediation for JavaScript and Go. Java and Python support will expand to cover newly disclosed vulnerabilities found through AI-assisted research, including flaws in dependencies coding agents may have pulled into a project. The changes build on Chainguard Repository, a policy-enforced endpoint for open source dependencies, and its Cursor partnership, which lets coding agents resolve packages through Chainguard's catalog instead of public registries.

Sources: The New Stack